A quality assessment of the soil water index by the propagation of ASCAT soil moisture error estimates through an exponential filter

Soil moisture data from remote-sensing measurements are used extensively and error characterization is critical in assessing their utility for specific applications. In hydrological monitoring and forecast applications, an estimate of the root-zone soil moisture is often required, and the exponential filter constitutes a common approach to obtaining a soil water index (SWI), i.e. the profile soil moisture from the remotely-sensed surface states. Advanced Scatterometer (ASCAT) surface soil moisture (SSM) time series include error estimates as ancillary data, however their effects are rarely considered in SWI computations. In this study, we introduced a simple error propagation scheme in parallel to exponential filter computation, aimed at estimating a SWI noise, which implicitly takes into account both errors and availability of the input SSM data actually used for each single SWI estimation. Integrating a control on SWI noise within quality check procedures was analysed by comparing in situ measurements available at different depths, up to 60 cm, for the period 2007–2014 in 10 stations located in Italy. The capacity of ASCAT-derived SWI to capture the same processes observed in the root-zone by in situ sensors, was evaluated by computing standard statistics (the correlation coefficient r and the root mean square difference [RMSD]) for different ‘a priori’ data masking procedures on SSM data, based on several indicators (processing flag, orbit direction, SSM noise) available for the ASCAT product. The effects of using the information contained in SSM noise and in the proposed SWI noise were also investigated. Removing data according to ‘a posteriori’ control on SWI noise rather than a priori masking SSM data, with the exponential filter T value and the percentage of removed data being equal, was found to generally improve the statistics of the resulting SWI time series. In addition, evaluating T on the SWI time series screened according to the SWI noise, outperforms other approaches analysed in almost every test case.